[This is a linkpost to my blog, cortrinkau.bearblog.dev. I regularly make art with DALL-E and have been making art with traditional media like pencil, ink, and watercolor for fifteen years. This article is a meditation on what makes AI art feel so different, after so long making art exclusively by hand.]
When I was a little kid, I loved to draw, and sometimes an idea for a drawing would come to me as I was falling asleep. At that moment, I felt inspired to make art, but spending the time to make the idea I had in my head come through on paper would have meant getting much less sleep, so I would write a description of the image in my head in order to still remember it the next morning. I still have notebooks and notebooks filled with art prompts for myself, and it’s fun to look through them and see what child-me wanted to draw---like “red boots walking on a telephone wire.”
Years later, that habit turned out to have been great practice at writing prompts for DALL-E. However, there’s a difference in what makes generating art with DALL-E feel satisfying versus the satisfaction of drawing it myself. When I have a specific image in my head, I want that exact image to be transferred to the paper, as faithfully as I can possibly get it. If my art skills aren’t quite up to what I want to draw, I’m often willing to spend time investing in my own drawing skills so I can get that image to look the way I want it. I feel frustrated when I can’t get the image to look the same as the image in my head, even if it still looks beautiful on its own.
Recently, I had a conversation with a friend recently about this, where I brought up the boots-on-telephone-wire example. As I was describing the image, I was generating an image in my own mind, but the image that those words generated in my friend’s mind was slightly different. When I try to generate that image with DALL-E, it wouldn’t get the specific details right – whether there’s a person wearing the boots, the angle the boots might be swaying at, the off-white texture of the paper, or the pencil-and-ink medium of the drawing, the Banksy-esque contrast of the black-and-white drawing and the hue of the girl’s boots. I have to draw the image myself, and I have to invest in my own drawing skills, if I want the exact image in my head to be out there in the world.
When I go to an art museum and look at oil paintings, I don’t think to myself, “I could never do that.” I think to myself about how many years it would take me to get to that level of skill. And I feel inspired to start that journey! Looking at oil paintings by Rembrandt, I feel inspired to learn oil painting and see if I can mimic any of the techniques I see in his paintings --- to see if I can get the smoothness of the lines or the delicate fading of light into shadow myself. It would feel so rewarding and deeply fulfilling to go on that journey. But that’s the problem with AI generation---instead of starting that journey, setting aside time to paint, buying oil paints and canvas and finding a space to work, now I think to myself “but... I can just tell DALL-E to make me the image I want as an oil painting, so there’s no reason to spend that much time learning to do this.” Even though taking the hard path would make me happier and more fulfilled. I do plenty of watercolor art, and it may be a similar destination that you reach, making a painting myself versus having DALL-E generate it for me, but the journey of making my own paintings is so deeply fulfilling and relaxing and good for me. I feel a deep connection to the work, some fundamental connection to my materials when I’m touching the paper and daubing the paint myself. It’s a very tender thing, putting brush to paper, and when I make art I do it very painstakingly, with attention to detail. That connection to your materials really isn’t there when the AI does it for you.
That’s the sad part about generative AI to me – it makes the path towards making art much easier, but it dissuades those who might have taken the hard path from taking the hard path, since the easy path is right there. But on a deeper level, the hard path is better in some way – it is what gives you the skills to make more art, which is where AI fails you. If you’ve trained and studied and have learned art, you can correct the AI’s mistakes – you can step in and draw hands with five fingers, draw people with normally shaped faces, if you so choose.
The thing about AI is there will always be gaps. There will always be places where the AI can’t quite do as good of a job as we’d like, and that is when it’s important for highly trained human experts to swoop in and do the job correctly. But if we become reliant on AI to do 90% of the work for us, we won’t have the skills to complete the last 10%. This is why, as humans, we have to invest in our own skills, which means studying and practicing and learning, so that if we ever lose this tool, or if we’re in a situation where AI can’t help us, we can go the rest of the way on our own. We must be able to carry ourselves forward on our own if we need to.